Tensor TownISSUE #96 of 120

speech recognition · whisper

NeuraVSThe Overfit Ogre

Neura saysSpeech recognition (Whisper) transcribes audio to text robustly across accents and noise.

Automatic speech recognition turns speech into text. OpenAI's Whisper (open-weight) set a strong bar: trained on diverse multilingual audio, it handles accents, background noise, and many languages, and can translate. It powers captions, voice interfaces, meeting notes, and the audio side of voice assistants. Run it locally or via API; it's the default building block for "voice in".

Power-ups you unlock

Audio → text (ASR)
Whisper: multilingual, noise-robust, open
Handles accents, can translate
Powers captions, voice UIs, notes

The Overfit Ogre attacks — common mistakes

Assuming perfect transcription in heavy noise
Ignoring language/locale settings
No handling of homophones/proper nouns

Boss battleList three products built on speech-to-text.

Example code

<!doctype html><html><head><meta charset="utf-8"></head>
<body style="background:#06040d;color:#e6e0ff;font-family:monospace;padding:20px"><pre>speech → Whisper → text
captions · meeting notes · voice assistant input
multilingual, noise-robust</pre></body></html>

▶ Open the interactive comic issue

‹ Text-To-Speech · Vits · Xtts · Elevenlabs Video Generation · Sora · Veo ›